veth.c revision cf05c700cf6dd6f28bd95586d3040f809fd365f5 
1/*
2 *  drivers/net/veth.c
3 *
4 *  Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc
5 *
6 * Author: Pavel Emelianov <xemul@openvz.org>
7 * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com>
8 *
9 */
10
11#include <linux/netdevice.h>
12#include <linux/slab.h>
13#include <linux/ethtool.h>
14#include <linux/etherdevice.h>
15#include <linux/u64_stats_sync.h>
16
17#include <net/dst.h>
18#include <net/xfrm.h>
19#include <linux/veth.h>
20
21#define DRV_NAME	"veth"
22#define DRV_VERSION	"1.0"
23
24#define MIN_MTU 68		/* Min L3 MTU */
25#define MAX_MTU 65535		/* Max L3 MTU (arbitrary) */
26
27struct veth_net_stats {
28	u64			rx_packets;
29	u64			tx_packets;
30	u64			rx_bytes;
31	u64			tx_bytes;
32	u64			rx_dropped;
33	u64			tx_dropped;
34	struct u64_stats_sync	syncp;
35};
36
37struct veth_priv {
38	struct net_device *peer;
39	struct veth_net_stats __percpu *stats;
40};
41
42/*
43 * ethtool interface
44 */
45
46static struct {
47	const char string[ETH_GSTRING_LEN];
48} ethtool_stats_keys[] = {
49	{ "peer_ifindex" },
50};
51
52static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
53{
54	cmd->supported		= 0;
55	cmd->advertising	= 0;
56	ethtool_cmd_speed_set(cmd, SPEED_10000);
57	cmd->duplex		= DUPLEX_FULL;
58	cmd->port		= PORT_TP;
59	cmd->phy_address	= 0;
60	cmd->transceiver	= XCVR_INTERNAL;
61	cmd->autoneg		= AUTONEG_DISABLE;
62	cmd->maxtxpkt		= 0;
63	cmd->maxrxpkt		= 0;
64	return 0;
65}
66
67static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
68{
69	strcpy(info->driver, DRV_NAME);
70	strcpy(info->version, DRV_VERSION);
71	strcpy(info->fw_version, "N/A");
72}
73
74static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
75{
76	switch(stringset) {
77	case ETH_SS_STATS:
78		memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
79		break;
80	}
81}
82
83static int veth_get_sset_count(struct net_device *dev, int sset)
84{
85	switch (sset) {
86	case ETH_SS_STATS:
87		return ARRAY_SIZE(ethtool_stats_keys);
88	default:
89		return -EOPNOTSUPP;
90	}
91}
92
93static void veth_get_ethtool_stats(struct net_device *dev,
94		struct ethtool_stats *stats, u64 *data)
95{
96	struct veth_priv *priv;
97
98	priv = netdev_priv(dev);
99	data[0] = priv->peer->ifindex;
100}
101
102static const struct ethtool_ops veth_ethtool_ops = {
103	.get_settings		= veth_get_settings,
104	.get_drvinfo		= veth_get_drvinfo,
105	.get_link		= ethtool_op_get_link,
106	.get_strings		= veth_get_strings,
107	.get_sset_count		= veth_get_sset_count,
108	.get_ethtool_stats	= veth_get_ethtool_stats,
109};
110
111/*
112 * xmit
113 */
114
115static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
116{
117	struct net_device *rcv = NULL;
118	struct veth_priv *priv, *rcv_priv;
119	struct veth_net_stats *stats, *rcv_stats;
120	int length;
121
122	priv = netdev_priv(dev);
123	rcv = priv->peer;
124	rcv_priv = netdev_priv(rcv);
125
126	stats = this_cpu_ptr(priv->stats);
127	rcv_stats = this_cpu_ptr(rcv_priv->stats);
128
129	if (!(rcv->flags & IFF_UP))
130		goto tx_drop;
131
132	/* don't change ip_summed == CHECKSUM_PARTIAL, as that
133	   will cause bad checksum on forwarded packets */
134	if (skb->ip_summed == CHECKSUM_NONE &&
135	    rcv->features & NETIF_F_RXCSUM)
136		skb->ip_summed = CHECKSUM_UNNECESSARY;
137
138	length = skb->len;
139	if (dev_forward_skb(rcv, skb) != NET_RX_SUCCESS)
140		goto rx_drop;
141
142	u64_stats_update_begin(&stats->syncp);
143	stats->tx_bytes += length;
144	stats->tx_packets++;
145	u64_stats_update_end(&stats->syncp);
146
147	u64_stats_update_begin(&rcv_stats->syncp);
148	rcv_stats->rx_bytes += length;
149	rcv_stats->rx_packets++;
150	u64_stats_update_end(&rcv_stats->syncp);
151
152	return NETDEV_TX_OK;
153
154tx_drop:
155	kfree_skb(skb);
156	u64_stats_update_begin(&stats->syncp);
157	stats->tx_dropped++;
158	u64_stats_update_end(&stats->syncp);
159	return NETDEV_TX_OK;
160
161rx_drop:
162	u64_stats_update_begin(&rcv_stats->syncp);
163	rcv_stats->rx_dropped++;
164	u64_stats_update_end(&rcv_stats->syncp);
165	return NETDEV_TX_OK;
166}
167
168/*
169 * general routines
170 */
171
172static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev,
173						  struct rtnl_link_stats64 *tot)
174{
175	struct veth_priv *priv = netdev_priv(dev);
176	int cpu;
177
178	for_each_possible_cpu(cpu) {
179		struct veth_net_stats *stats = per_cpu_ptr(priv->stats, cpu);
180		u64 rx_packets, rx_bytes, rx_dropped;
181		u64 tx_packets, tx_bytes, tx_dropped;
182		unsigned int start;
183
184		do {
185			start = u64_stats_fetch_begin_bh(&stats->syncp);
186			rx_packets = stats->rx_packets;
187			tx_packets = stats->tx_packets;
188			rx_bytes = stats->rx_bytes;
189			tx_bytes = stats->tx_bytes;
190			rx_dropped = stats->rx_dropped;
191			tx_dropped = stats->tx_dropped;
192		} while (u64_stats_fetch_retry_bh(&stats->syncp, start));
193		tot->rx_packets += rx_packets;
194		tot->tx_packets += tx_packets;
195		tot->rx_bytes   += rx_bytes;
196		tot->tx_bytes   += tx_bytes;
197		tot->rx_dropped += rx_dropped;
198		tot->tx_dropped += tx_dropped;
199	}
200
201	return tot;
202}
203
204static int veth_open(struct net_device *dev)
205{
206	struct veth_priv *priv;
207
208	priv = netdev_priv(dev);
209	if (priv->peer == NULL)
210		return -ENOTCONN;
211
212	if (priv->peer->flags & IFF_UP) {
213		netif_carrier_on(dev);
214		netif_carrier_on(priv->peer);
215	}
216	return 0;
217}
218
219static int veth_close(struct net_device *dev)
220{
221	struct veth_priv *priv = netdev_priv(dev);
222
223	netif_carrier_off(dev);
224	netif_carrier_off(priv->peer);
225
226	return 0;
227}
228
229static int is_valid_veth_mtu(int new_mtu)
230{
231	return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
232}
233
234static int veth_change_mtu(struct net_device *dev, int new_mtu)
235{
236	if (!is_valid_veth_mtu(new_mtu))
237		return -EINVAL;
238	dev->mtu = new_mtu;
239	return 0;
240}
241
242static int veth_dev_init(struct net_device *dev)
243{
244	struct veth_net_stats __percpu *stats;
245	struct veth_priv *priv;
246
247	stats = alloc_percpu(struct veth_net_stats);
248	if (stats == NULL)
249		return -ENOMEM;
250
251	priv = netdev_priv(dev);
252	priv->stats = stats;
253	return 0;
254}
255
256static void veth_dev_free(struct net_device *dev)
257{
258	struct veth_priv *priv;
259
260	priv = netdev_priv(dev);
261	free_percpu(priv->stats);
262	free_netdev(dev);
263}
264
265static const struct net_device_ops veth_netdev_ops = {
266	.ndo_init            = veth_dev_init,
267	.ndo_open            = veth_open,
268	.ndo_stop            = veth_close,
269	.ndo_start_xmit      = veth_xmit,
270	.ndo_change_mtu      = veth_change_mtu,
271	.ndo_get_stats64     = veth_get_stats64,
272	.ndo_set_mac_address = eth_mac_addr,
273};
274
275static void veth_setup(struct net_device *dev)
276{
277	ether_setup(dev);
278
279	dev->netdev_ops = &veth_netdev_ops;
280	dev->ethtool_ops = &veth_ethtool_ops;
281	dev->features |= NETIF_F_LLTX;
282	dev->destructor = veth_dev_free;
283
284	dev->hw_features = NETIF_F_NO_CSUM | NETIF_F_SG | NETIF_F_RXCSUM;
285}
286
287/*
288 * netlink interface
289 */
290
291static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
292{
293	if (tb[IFLA_ADDRESS]) {
294		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
295			return -EINVAL;
296		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
297			return -EADDRNOTAVAIL;
298	}
299	if (tb[IFLA_MTU]) {
300		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
301			return -EINVAL;
302	}
303	return 0;
304}
305
306static struct rtnl_link_ops veth_link_ops;
307
308static int veth_newlink(struct net *src_net, struct net_device *dev,
309			 struct nlattr *tb[], struct nlattr *data[])
310{
311	int err;
312	struct net_device *peer;
313	struct veth_priv *priv;
314	char ifname[IFNAMSIZ];
315	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
316	struct ifinfomsg *ifmp;
317	struct net *net;
318
319	/*
320	 * create and register peer first
321	 */
322	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
323		struct nlattr *nla_peer;
324
325		nla_peer = data[VETH_INFO_PEER];
326		ifmp = nla_data(nla_peer);
327		err = nla_parse(peer_tb, IFLA_MAX,
328				nla_data(nla_peer) + sizeof(struct ifinfomsg),
329				nla_len(nla_peer) - sizeof(struct ifinfomsg),
330				ifla_policy);
331		if (err < 0)
332			return err;
333
334		err = veth_validate(peer_tb, NULL);
335		if (err < 0)
336			return err;
337
338		tbp = peer_tb;
339	} else {
340		ifmp = NULL;
341		tbp = tb;
342	}
343
344	if (tbp[IFLA_IFNAME])
345		nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
346	else
347		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
348
349	net = rtnl_link_get_net(src_net, tbp);
350	if (IS_ERR(net))
351		return PTR_ERR(net);
352
353	peer = rtnl_create_link(src_net, net, ifname, &veth_link_ops, tbp);
354	if (IS_ERR(peer)) {
355		put_net(net);
356		return PTR_ERR(peer);
357	}
358
359	if (tbp[IFLA_ADDRESS] == NULL)
360		random_ether_addr(peer->dev_addr);
361
362	err = register_netdevice(peer);
363	put_net(net);
364	net = NULL;
365	if (err < 0)
366		goto err_register_peer;
367
368	netif_carrier_off(peer);
369
370	err = rtnl_configure_link(peer, ifmp);
371	if (err < 0)
372		goto err_configure_peer;
373
374	/*
375	 * register dev last
376	 *
377	 * note, that since we've registered new device the dev's name
378	 * should be re-allocated
379	 */
380
381	if (tb[IFLA_ADDRESS] == NULL)
382		random_ether_addr(dev->dev_addr);
383
384	if (tb[IFLA_IFNAME])
385		nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
386	else
387		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
388
389	if (strchr(dev->name, '%')) {
390		err = dev_alloc_name(dev, dev->name);
391		if (err < 0)
392			goto err_alloc_name;
393	}
394
395	err = register_netdevice(dev);
396	if (err < 0)
397		goto err_register_dev;
398
399	netif_carrier_off(dev);
400
401	/*
402	 * tie the deviced together
403	 */
404
405	priv = netdev_priv(dev);
406	priv->peer = peer;
407
408	priv = netdev_priv(peer);
409	priv->peer = dev;
410	return 0;
411
412err_register_dev:
413	/* nothing to do */
414err_alloc_name:
415err_configure_peer:
416	unregister_netdevice(peer);
417	return err;
418
419err_register_peer:
420	free_netdev(peer);
421	return err;
422}
423
424static void veth_dellink(struct net_device *dev, struct list_head *head)
425{
426	struct veth_priv *priv;
427	struct net_device *peer;
428
429	priv = netdev_priv(dev);
430	peer = priv->peer;
431
432	unregister_netdevice_queue(dev, head);
433	unregister_netdevice_queue(peer, head);
434}
435
436static const struct nla_policy veth_policy[VETH_INFO_MAX + 1];
437
438static struct rtnl_link_ops veth_link_ops = {
439	.kind		= DRV_NAME,
440	.priv_size	= sizeof(struct veth_priv),
441	.setup		= veth_setup,
442	.validate	= veth_validate,
443	.newlink	= veth_newlink,
444	.dellink	= veth_dellink,
445	.policy		= veth_policy,
446	.maxtype	= VETH_INFO_MAX,
447};
448
449/*
450 * init/fini
451 */
452
453static __init int veth_init(void)
454{
455	return rtnl_link_register(&veth_link_ops);
456}
457
458static __exit void veth_exit(void)
459{
460	rtnl_link_unregister(&veth_link_ops);
461}
462
463module_init(veth_init);
464module_exit(veth_exit);
465
466MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
467MODULE_LICENSE("GPL v2");
468MODULE_ALIAS_RTNL_LINK(DRV_NAME);
469